Graphify: code knowledge graph#

Graphify builds a queryable graph of source code. Mantid’s configuration limits the graph to the framework and GUI code: Framework/ and scripts/, which make up the mantid conda package, and qt/, which makes up the mantidqt and mantidworkbench packages. Documentation and tests are not included.

The graph holds classes, functions, includes and imports, calls and references, grouped into communities of closely connected code. It is useful for questions such as “what is connected to MatrixWorkspace?” or “which parts of the GUI in qt/ use the Python API in Framework/?”.

Mantid configures graphify for a code-only run. Code is parsed locally with tree-sitter, so building the graph makes no LLM calls and needs no API key. The one exception is naming communities, described under Usage.

Installing#

The name of the package is graphifyy (note the double y) and is not part of the developer environment. It can be installed pixi add --global graphifyy then it can be used inside pixi shell or through pixi run graphify. See the installation guide for other ways to install it.

Configuration#

Two files in the root of the repository control what goes into the graph.

.graphifyignore

Uses gitignore-style patterns. It limits the graph to Framework/, scripts/ and qt/. Within those folders it also excludes tests, documentation, images, and generated or vendored code (gSOAP for ICat, FlatBuffers for LiveData, the Qt Solutions property browser and generated Qt resources).

.graphifyrc

Holds project settings. It tells graphify to parse .h and .hxx files as C++.

Usage#

See the CLI quickstart and the CLI reference for all commands and options.

AI coding assistants#

graphify install adds a /graphify skill to coding assistants such as Claude Code, which can then answer questions from the graph. See using graphify with coding assistants. Contributions made with an AI assistant must follow the AI policy.

Limitations#

  • Algorithms are created by name (AlgorithmManager::Instance().create("Name")), so those calls do not appear as edges in the graph. Use grep to find them.

  • Names are matched without their namespace. As a result, std::string references are attached to an unrelated string enum value, which then appears as the most connected node.

  • Graphify does not run the C++ preprocessor. Mantid’s macros (DECLARE_ALGORITHM, PARALLEL_FOR_IF and the export macros) therefore cause parse errors in many files, but very few symbols are lost.

  • The graph is a snapshot of the code when it was built. Run graphify update . after changing code, or install the git hooks.